Ask Slashdot: Safe Learning Environment For VMs?
First time accepted submitter rarkian writes "I am the teacher in this story. I teach Python and C++ to high school students: grades 9-12. I use CentOS 6 with DRBL to run my computer lab. Some of my students have become Linux experts. Next year I'm planning on allowing students to create and run their own VMs in a segregated LAN. Any advice on which virtualization technology to use and security concerns with allowing students to be root in a VM?"
for each of the students and don't allow any interface between them...and certainly no main network/internet access.
Vagrant is a wrapper for Virtualbox and VMWare Workstation that accelerates the deployment of development environments.
http://www.vagrantup.com/
Unless they take down the network, e.g. running a rogue DHCP server. Or they use it to hack other systems on the network, e.g. password-sniffing the other student's credentials when they log in from their VMs.
if the VM has a full root account, with a network address on the global network at large, then it has the ability to, for example, run a priviledged NMAP scan on the entire network. Which can expose open ports or vulnerabilities on another machine that can then be used to leverage access.
The fact that your using VM's is largely moot and goes back to the line of thought that VM's are somehow not 'real' computers. VM's run the same operating systems, software, have the same bugs, vulnerabilities and everything else as a physical computer. You need to patch them just like any other computer and you need to license them just like a regular computer. The fact that they are VM's really only makes two differences practical differences that matter, fist is that is easy to roll them back and second is that your aren't running on bare metal.
In other words you have a core issue that needs addressed of giving students root access to a computer. In an isolated environment this isn't necessarily a bad thing. Understand that they exploit root and see what they can do with it, however they are there to learn and if you can do so safely and without disruption of what your trying to teach then let them. Your focus needs to be on making it safe for those around them and that means making sure your VLAN and any related Internet access are properly setup. The lab is a lab and as long as you can make sure they aren't getting access to anyone persons computer than let them have at it.
A good rule of thumb is to roll your sessions back prior to the start of every single class. This always gives a fresh machine and the students will quickly learn how to set their VM just the way they want it.
Make sure you have SELinux enabled (and enforcing!) on the VM host, and keep the VMM software updated... there sometimes are security holes in VMM software which can be exploited. SELinux can help contain a breached VMM.
Just in case anyone gets a bit... shall we say "Adventurous" and tries to use their root access boxen to attack something they shouldn't, it might be worth isolating the VMs on their own VLAN away from the rest of the network, if you haven't already.
Next year I'm planning on allowing students to create and run their own VMs
Running their own VM's is straightforward. Allowing the students to create their own VM's implies that they'll be root on the hypervisor.
Do you intend to run the hypervisor on the client machines of the DRBL system, or run a single hypervisor on the server and deploy the VM's there as DRBL clients?
To satisfy your requirements you probably want to run the hypervisor on the clients so they students can each have their own root on the hypervisor. This would require a hypervisor compatible with DRBL. I don't know how it works, but just from reading the description on the webpage, it sounds like it's geared to PXE booting a host OS.
If you go with Xen, you'll have to probably separately PXE boot Xen and then DRBL boot the Dom0. Which would probably work fine and get you decent performance, but it will expose the students to DRBL (is this what you want?)
If you go with KVM, the performance is a bit slower, but for a student shop that's probably OK, and you'll be able to DRBL-deploy the hypervisor and then let the students create their own non-DRBL (or DRBL) guests. This probably fits your model the best unless you have old hardware that KVM does not support - then you might need to go with the Xen-PXE-Boot model (because it can paravirtualize without hardware assistance).
You could also use VirtualBox, and while it offers a nice GUI, it's probably too simple for teaching your students about virtualization (it just feels like an app).
BTW, it sounds like you're doing great work based on that article. Kudos on your accomplishments and being an inspiration for others in your field.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
And is utter trash for anything that needs to be scalable.
It is fine for a desktop VM system, but it simply does not offer the management interfaces that other solutions have. Basically the options here are VMware and KVM. The first if you want a shiny GUI the latter if you are ok without one. Both will let you script everything they do, which will be very handy when you need to reset 100 VMs for the next batch of students.
I see this as being similar to when we needed to have all of our developers in my company working in an environment that absolutely matched the production environment. Just use VMWare on each individual machine, run an Ubuntu image in that and best of all use a Puppet script to customize it and give 'em the goodies they need. The beauty of this is once the kids screw it up (and let's hope they do, they're learning after all) then you can rebuild this back to a pristine machine in no time. Good luck!
They can always plug in their own laptop and do that anyway.
Depending on your equipment and the time you want to spend, oVirt might be an answer.
Although it is still fairly new and is in development, it runs on CentOS6, is free, can handle multiple guest OSes, can create VM's from a template, and has a power users portal page where trusted students/employees can create their own VM from supplied templates. This way, no student would have access to the host OS, but could create a VM as needed. The downside is that it can get quite complicated to set up the system, and could take a bit of time to learn and set it up properly. Since it is free, you are also dependent upon community support.
You can access more info here.
Great civilizations have lived and died on false theories. Don't mess up mine with a few facts.
Which is why they need to setup their own VLAN to isolate the VMs to the classroom. VM traffic is isolated to non-routing VLANS. They call this setup a "sandbox", and it is generally a good practice for classroom work.
As for which VM technology to use ... VMWare, or ZEN or even Microsoft's version are usable. VMWare is sort of free, Xen definitely is. I'm not familiar with pricing on Microsoft's versions but schools tend to get steep discounts for server licenses. Look at OpenStack for management, I hear it is decent when it works.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
ESXi. Its free, powerful, and offers a lot of pre-built appliances. I don't see any safety concerns if the network is segregated. If you have specific VM's that you want the students to learn within, keep screenshots of those so that you may roll anything back that gets damaged. This is great because it allows them do pretty much anything they want without creating a maintenance headache for you.
If you want to teach them about specific technologies using VMs that go hand and hand with programming (like source control, bugzilla, configuring web servers, etc), turnkeylinux.org offers many free linux appliances that will make your job easy.
Meow
I don't think you need to worry about OS security, since that is the point of using VMs. However, the "key" to this question is the definition of "segmented." There are host of nefarious and simple mistakes you can make to completely trash the network of the of the VMs. I would recommend disabling multicast.
Banning the use of fork() can't hurt either.
Yes, banning fork() can hurt, because how else are you supposed to learn about it. Also, running a forkbomb in a VM would have no effect at all on the VM host.
Thanks for going the extra mile with your students.
As AC said, a separate LAN or VLAN, or multiple separate LANs/VLANs handles most of what's posted below. For example, a rogue DHCP server would only be visible on that VLAN.
Red Hat has a Virtualization Security section in their manual:
https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Virtualization_Administration_Guide/chap-Virtualization-Security_for_virtualization.html
CentOS/RHEL includes comprehensive support for KVM with virt-manager. While VirtualBox et al are fine for running one or two virtual machines on your desktop, for many VMs, with new ones created and removed each semester, the enterprise level support of KVM built into the distro is more appropriate. That support includes creating VLANs within the same management interface, for example, and integrates with the built in storage stack administration tools. Again, VirtualBox may be simpler to set up for one or to two machines, so I'm not saying it's not good - it's just not the best tool in this particular scenario. In this type of scenario, the KVM / virt-manager / virsh stack that RH baked in is probably a better match to the needs.
Forkbomb is only successful if you don't have limits on your VM environment. You have put limits on your environment, right?
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
A nice forkbomb in a single VM can cause headaches for the rest of the environment.
Then it's a very poor environment.
We're talking about one or many classes of students. If it's not built out to handle several VMs using their max CPU concurrently, then it's a very poor environment.
Heck, everyone compiling at the same time would shut things down if the environment is built poorly.
There are two types of people in the world: Those who crave closure
They can start each class with a fresh snapshot. In effect they would be restoring from backups. The configuration files from some other networked storage or their thumb drives and the applications themselves from the repositories. I've done something similar, but on bare metal, and after about half a dozen times they don't notice -- it had become such second nature to install and restore applications. Heck you might even have them practice installing the whole system from scratch. If you go that route, they can become quite proficient with installation and resource allocation. PXE booting a netinstall image helps there.
However, once you start to load packages from the net things can really slow down unless you prepare. The best way is to have a cache like APT-Cacher or Squid on your LAN or host system and have them configure their systems to use it for APT. For the cache to be most effective, you have to pre-load it before each class. That's easy and can be done while doing other things. It only takes time not attention. But once you have the cache loaded, installation will fly and can be done in 15 - 20 minutes. After that they weren't shy about installing on their own computers at home or helping their friends.
Beta is broken and the link to classic doesn't work. Stop wasting our time or there won't be anybody left here.
So far, I see lots of advice about VM breakouts and network isolation. If this were a production datacenter where uptime was a criteria, this is all well and good. I suspect that this isn't what you need to hear, however.
I see three things you could be attempting to protect:
1) The larger school network.
2) The VM host infrastructure.
3) The VMs themselves.
1) A student on a VM is no more dangerous to the network than one who can connect to the school wireless with a laptop or smartphone. If the lab uplinks to the same network as the broader access, your risk profile is unchanged.
2) Make sure the VMs can't route to the host and keep it patched. If a student managed to break out of a VM in a patched hosting environment, do some forensics and find the bug then sell it. It's probably worth more than you make in a year. Seriously, if they can do this, they deserve to win. You might as well worry about protecting against nation-state sponsored attacks.
3) Make sure that the class work is backed up (a git server, perhaps) and then don't worry about it. Seriously, just throw the VMs away after each class (or every night, etc) and start with a clean one the next time they log in. Don't spend time trying to outsmart a classroom full of bored highschoolers. Instead, make it so it doesn't matter when they break something.
Wow, two whole servers. However did you scale so high?
Unless they take down the network, e.g. running a rogue DHCP server. Or they use it to hack other systems on the network, e.g. password-sniffing the other student's credentials when they log in from their VMs.
So... nothing they couldn't do much easier/more safely by just pulling the network cable out of the physical machine and connecting it to their netbook?
No sig today...
You know how I know you didn't even read the summary...?
No sig today...
I am not familiar with KVM or other Linux VM solutions.
I do know that during my VCP cert course, all students were provided with a VMWare ESXi infrastructure that was entirely virtual, contained in a vApp on a parent vSphere infrastructure. We all had our own connection on our own vSwitch, but no uplinks to everyone else, so there really wasnt much anyone could do to interfere with other students.
I suppose one of the students could try to defeat the vSwitch segregation via an exploit, but I think if they pull that off they dont need the class and deserve an instant A (assuming responsible disclosure).
VirtualBox provides the VBoxManage tool for automating operations. It works perfectly fine for this sort of thing. One of my small servers at home is running 28 VMs with all management happening through the command line, and that hasn't even gotten close to whatever the upper limit is. You certainly can run a classroom worth of VMs on a modestly sized box.
The only major management feature that's much easier on VMWare than VirtualBox is moving VMs to new systems. That is very useful for large production VM deployments, but it doesn't sound necessary for this situation.
I use VirtualBox because there is an open source release that works across multiple platforms. VMWare is all closed, and Xen only works on UNIX-ish systems. Students in particular can benefit from running a VM copy of Linux on another host OS, because it provides a way to get familiar with the software on either a Windows or Mac laptop (which they probably own already). That requirement rules out Xen as a good example. And if you're going to introduce students to open source software via Linux, it's nice if you can present that lesson on an open source stack too.
So - have a lesson on forkbombs prepared for when that happens.
retrorocket.o not found, launch anyway?
So, I co-wrote this book on virtual security and am a former VMware Cloud Solutions Architect. And I'll preface this advice by saying that, if you want to talk more in depth, feel free to ping me. First initial, last name at gmail will work. (The email I have attached to slashdot I glance at occasionally, but it gets almost purely spam and so I'd likely miss anything.)
From my perspective, the first question is which hypervisor to use:
- VMware is mature, you can get a free license for the base hypervisor (which is quite feature rich; this is no trial product) for up to 32GB per physical box, is widely used. If VMware remains as relevant in the future as it is now, it's actually a very solid skillset to have.
- If you have physical hosts over 32GB, VMware ceases to be free
- Some features require more advanced VMware stuff, including vCenter server, which isn't free - for example, VMware's live vm migration feature (vMotion)
- VMware is almost entirely closed on the internals; hypervisor is closed source (other than a not-useful-for-your-purposes "open source" bundle that contains their modified GPL code only); they have a bunch of APIs for internal functions (ie, tracking changed blocks on the virtual iscsi devices, for example), but those are generally restricted to partners; so if your students want to actually hack the virtualization layer, they can't. Then again, letting them do so wouldn't really be safe.
- On the other hand, VMware layers do have nice APIs that are reasonably accessible for doing non-internals stuff; things like powering VMs on and off, changing their allocated RAM and cpus, etc
- VMware has a nice set of tools, including CLI tools, which work well even with the free versions, that can allow you to move virtual machines in and out of specific hypervisors (not while the VMs are powered on), and into and out of VMware's desktop products (Workstation for Windows and Linux, Fusion for Mac). (google ovftool for the cross-platform CLI tool, for example; it can import/export to/from ESX, vCenter Server, Workstation, Fusion, and vCloud instances)
- VMware has a nice set of tools for snapshots and backups, even on the base hypervisor; for example, I have a personal ESX box at a provider and I use this tool to back up the VMs back and forth, which can be done from outside the OS without powering the VM down, and it's free.
- I found using some things I'd think of as mandatory for a lab environment (ie, thin provisioning) were just built-in on the VMware side and required a fair bit of extra work and added extra wrinkles
The virtual networking on VMware is dramatically more mature from my experience; my experience with Xen & KVM is now dated (it's been 2 years since I was in the thick of writing that book, which was the last time I was really in the thick of exploring the open-source hypervisor networking bits). I found that depending on the version of the hypervisor OS, which hypervisor, which kernel, which guest, etc, you could fall into all sorts of traps. I had some examples in the book where I showed, for example, generating and applying ebtables configurations to the host OS (the Xen Linux hypervisor OS) to block forged frames from coming across the bridge from one of the guest Linuxes, for example.
Compare that to the VMware side, you could in theory wire up everything to dumb hubs, even, and enforce network separation at the hypervisor layer with VLAN tags applied to the portgroups where you attach VMs. (Warning: not suggesting you blindly do that; but VLAN enforcement on the VMware side is fairly rigid if configured in a good way.)
My own book is a fun read for some of these concerns, although Haletky's book is probably the canonical work on the subject. (Although it is -slightly- dated from bein